COMPMID-1050 CL/NEON: Create a function to convert the 2D weights of FC layer from NHWC to NCHW and viceversa

Change-Id: If77ffeb92b6eb883e5d2d2c97c2c4d1d23d17c8d
Reviewed-on: https://eu-gerrit-1.euhpc.arm.com/129257
Tested-by: Jenkins <bsgcomp@arm.com>
Reviewed-by: Gian Marco Iodice <gianmarco.iodice@arm.com>
Reviewed-by: Anthony Barbier <anthony.barbier@arm.com>

1 files changed, 81 insertions, 0 deletions

diff --git a/arm_compute/core/CL/kernels/CLConvertFullyConnectedWeightsKernel.h b/arm_compute/core/CL/kernels/CLConvertFullyConnectedWeightsKernel.h
new file mode 100644
index 0000000000..fe24aa9d8c
--- /dev/null
+++ b/arm_compute/core/CL/kernels/CLConvertFullyConnectedWeightsKernel.h
@@ -0,0 +1,81 @@
+/*
+ * Copyright (c) 2018 ARM Limited.
+ *
+ * SPDX-License-Identifier: MIT
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to
+ * deal in the Software without restriction, including without limitation the
+ * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
+ * sell copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#ifndef __ARM_COMPUTE_CLCONVERTFULLYCONNECTEDWEIGHTSKERNEL_H__
+#define __ARM_COMPUTE_CLCONVERTFULLYCONNECTEDWEIGHTSKERNEL_H__
+
+#include "arm_compute/core/CL/ICLKernel.h"
+
+namespace arm_compute
+{
+class ICLTensor;
+
+/** Interface to convert the 2D Fully Connected weights from NCHW to NHWC or vice versa.
+ *
+ * @note This function can be applied to the 2D weights used by a Fully Connected layer if:
+ *       - It follows a Convolution layer
+ *       - The data layout used by the network does not match the one the model has been trained in.
+ *
+ * @note This function assumes the weights are already reshaped (transposed)
+ */
+class CLConvertFullyConnectedWeightsKernel : public ICLKernel
+{
+public:
+    /** Default constructor */
+    CLConvertFullyConnectedWeightsKernel();
+    /** Prevent instances of this class from being copied (As this class contains pointers) */
+    CLConvertFullyConnectedWeightsKernel(const CLConvertFullyConnectedWeightsKernel &) = delete;
+    /** Prevent instances of this class from being copied (As this class contains pointers) */
+    CLConvertFullyConnectedWeightsKernel &operator=(const CLConvertFullyConnectedWeightsKernel &) = delete;
+    /** Allow instances of this class to be moved */
+    CLConvertFullyConnectedWeightsKernel(CLConvertFullyConnectedWeightsKernel &&) = default;
+    /** Allow instances of this class to be moved */
+    CLConvertFullyConnectedWeightsKernel &operator=(CLConvertFullyConnectedWeightsKernel &&) = default;
+    /** Default destructor */
+    ~CLConvertFullyConnectedWeightsKernel() = default;
+    /** Set the input and output tensor.
+     *
+     * @param[in]  input                Source weights tensor to convert. Must be 2 dimensional. Data types supported: U8/S8/QS8/QASYMM8/U16/S16/QS16/U32/S32/QS32/F16/F32.
+     * @param[out] output               The converted weights tensor. Shape and Data Type: Same as @p input.
+     * @param[in]  original_input_shape Shape of the original input tensor (the one entering fully connected layer). Must be in NCHW format.
+     * @param[in]  data_layout          The data layout the weights have been trained in.
+     */
+    void configure(const ICLTensor *input, ICLTensor *output, const TensorShape &original_input_shape, DataLayout data_layout);
+    /** Static function to check if given info will lead to a valid configuration of @ref CLConvertFullyConnectedWeightsKernel
+     *
+     * @param[in] input                Source weights tensor info to convert. Must be 2 dimensional. Data types supported: U8/S8/QS8/QASYMM8/U16/S16/QS16/U32/S32/QS32/F16/F32.
+     * @param[in] output               The converted weights tensor info. Shape and Data Type: Same as @p input.
+     * @param[in] original_input_shape Shape of the original input tensor (the one entering fully connected layer). Must be in NCHW format.
+     * @param[in] data_layout          The data layout the weights have been trained in.
+     */
+    static Status validate(const ITensorInfo *input, const ITensorInfo *output, const TensorShape &original_input_shape, DataLayout data_layout);
+
+    // Inherited methods overridden:
+    void run(const Window &window, cl::CommandQueue &queue) override;
+
+private:
+    const ICLTensor *_input;
+    ICLTensor       *_output;
+};
+} // namespace arm_compute
+#endif /*__ARM_COMPUTE_CLCONVERTFULLYCONNECTEDWEIGHTSKERNEL_H__ */